Fraud detection in healthcare reimbursement is critical for safeguarding the integrity of financial transactions between hospitals and insurance companies. While traditional machine learning techniques have been widely used for this purpose, they face challenges in capturing the complex relationships inherent in insurance networks, scalability issues, and handling imbalanced datasets. To address these limitations, this study proposes a novel approach using Graph Neural Networks (GNNs) and graph analysis for insurance fraud detection. The primary objective is to compare the effectiveness of traditional machine learning methods with GNNs in identifying fraudulent activities in hospital insurance interactions. By representing the relationships between hospitals and insurance companies as a graph, the study examines the ability of GNNs to detect intricate fraud patterns that may be overlooked by traditional methods. Real-world datasets from Medicare claims are employed to evaluate the performance, scalability, and practical implications of both approaches. GNNs offer several advantages over traditional machine learning models. They excel in capturing the complex relationships within insurance networks, providing a more comprehensive understanding of fraudulent activities. Unlike traditional models, GNNs inherently grasp interconnected relationships between entities, enabling them to adapt dynamically to evolving fraud schemes. Moreover, GNNs demonstrate enhanced contextual awareness by analysing the entire network to identify fraudulent patterns that may be obscured in isolated transactions. These advantages position GNNs as a promising technology for enhancing fraud detection in the healthcare sector. However, the adoption of GNNs presents challenges. Implementing GNNs requires a deeper level of technical expertise and computational resources compared to traditional machine-learning models. Additionally, the complexity of GNN architectures may lead to overfitting, necessitating specialized techniques such as dropout and mini-batch training to mitigate these issues. Despite these challenges, the potential benefits of GNNs in detecting fraud justify the exploration of this advanced technology. This study adopts a modular approach to investigate insurance fraud detection, encompassing modules such as dataset description, fraud detection model development, graph construction, and logistic regression. Through these modules, the research aims to provide a comprehensive understanding of the processes involved in implementing GNNs for fraud detection. In conclusion, this research contributes to the growing body of knowledge in the field of insurance fraud detection by exploring the potential of Graph Neural Networks. By comparing traditional machine learning methods with GNNs, this study sheds light on the advantages and challenges associated with adopting advanced technologies for fraud detection in critical sectors such as healthcare reimbursement.